Parallel Hyperparameter Tuning of Accuracy for Deep Learning based Tornado Predictions
MetadataShow full item record
Type of Work9 pages
Citation of Original PublicationJonathan N. Basalyga, Parallel Hyperparameter Tuning of Accuracy for Deep Learning based Tornado Predictions, http://hpcf-files.umbc.edu/research/papers/Basalyga_SeniorThesis2020.pdf
RightsThis item is likely protected under Title 17 of the U.S. Copyright Law. Unless on a Creative Commons license, for uses protected by Copyright Law, contact the copyright holder or the author.
Predicting violent storms and dangerous weather conditions with current physics based weather models can take a long time due to the immense complexity associated with numerical simulations. Machine learning has the potential to classify tornadic weather patterns much more rapidly, thus allowing for more timely alerts to the public. In this work, we examine what impact varying the batch size and the number of GPUs a convolutional neural network is trained on has on the network’s accuracy at classifying storm data. We conclude that using multiple GPUs to train a single network has no significant advantage over using a single GPU. Therefore, multiple GPUs should instead be used to maximize search throughput by using each of them simultaneously for single GPU runs or to solve larger problems by pooling their memory.